Agent Examples
Practical examples showing how AI agents work with a codex vault. Each example demonstrates a different agent platform connecting to the same underlying structure.
The codex is agent-agnostic — any system that can read markdown files and follow conventions can use it.
Examples
| File | Description |
|---|---|
| openclaw-workspace.md | How an OpenClaw agent workspace connects to and operates on a codex |
| letta-context-repo.md | Using a codex as a Letta agent’s Context Repository with memory tiers |
| cursor-rules.md | Setting up Cursor, Claude Code, or Copilot to work with a codex |
How Agents Use The Codex
Every agent integration follows the same core pattern:
- Boot — Read
_codex/system/for pinned context (identity, conventions, domain map) - Search — Index domain folders for semantic retrieval
- Execute — Follow procedures from
_codex/skills/ - Write — Capture new knowledge back into the codex using conventions
The specifics vary by platform, but the structure stays the same.